[gmx-users] Constraints not working in pull code (sometimes, sometimes not)

[chris.neale at utoronto.ca]

Dear Krapnik:

1. please make the test cases identical, that means doing a simulation  
that you may not really be interested in so that the solutes are the  
same for both lipid and octane. It also means setting the  
displacements to similar values in my opinion (because perhaps the  
problem is based on your box size along z and choice of pull_pbcatom  
for example). There is *something* strange happening, but let's  
eliminate all other possibilities before assuming that there is a  
problem with the source code.

2. Once you have a setup in which the only difference is octane vs.  
lipid (really the only difference please -- even add your octane  
pressure settings to the lipid simulation) and you find strange  
behaviour for octane and not lipid, then please attempt that same  
system with pull=umbrella in place of pull=constraints and see if the  
solute also drifts far from its initial position.

3. Your initial definition of the problem was a little misleading.  
Both Justin and I focused on third decimal place differences, but now  
you are saying that there is nm-scale motion along z. Please report  
those values for your test systems, which are outlined above, after  
you redo your tests with identical settings other than the switch from  
lipid to octane. -- I no longer thing that a double precision test is  
necessary if you are seeing nm-scale movement along z.

4. Perhaps I was a little hasty to complain about a bug-report style  
post. I did like all of the information that you provided. I just got  
the impression from your title and the end of your post that you  
thought it was a problem with gromacs (which it may still be) and I  
think that this is generally a counter-productive attitude because it  
hinders one from making the proper tests.

Chris

-- original message --

Re: Constraints not working in pull code (sometimes, sometimes not)
Krapnik krapnik at gmail.com
Thu Aug 11 09:53:24 CEST 2011

     * Previous message: [gmx-users] append files in Gromacs 3.3
     * Next message: [gmx-users] Constraints not working in pull code  
(sometimes, sometimes not)
     * Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

Dear Justin
Thank you for your responses, but it still is not clear to me why is that
happening

> I don't see any real problem here.  The DOPC membrane probably holds its
shape
> better than the octanol slab and thus the reference position (which
determines
> the constraint) is less mobile.

That is certain. When we were running free simulation, the box started to
narrow. Therefore we have frozen the box in xy plane.

> As such, the constraint is more stable in DOPC
> than in octanol.  The slight bump in the octanol system amounts to a
change of
> 0.16%, which I'd say is quite small.

The trouble is, that during next 10 ns the distance was between 2.0 - 4.0
and molecule moved outside of the ocatnol slab and inside, which is then
useless for calculation of PMF.

> You're also looking at the first few
> frames of the simulation, which may amount to equilibration, but you
haven't
> mentioned if you've done anything prior to this run.

Previously, a free simulation with molecule on slab was run for 20 ns, from
which frame with small molecule at specidied depth was taken, so I suppose,
that the start is equilibrated.

> As long as there is no systematic change in position, I'd say there's no
problem.

Unfortunately, there is as I have said before.

-Justin

Dear Chris,
thank you for responses as well.

> I agree that Justin is probably correct, although constraints should
> technically work just fine with a highly dynamic reference group. Any
> problems should show up as the system blowing up, but not in correctly
> setting the position.

I hope, that the position is set up ok, as it stays steady in DOPC.

> I think that the problems can arise, however,
> when you have multiple competing constraints. You might, for example,
> try without constraining the octane bonds -- I think that you should
> get perfect match in that case.

I will try.
The erroneous movement of molecule seems to be enhanced within XY-plane
frozen.
The curious thing is however, that when I did not constrain size of
xy-plane, then the distance to the COM of slab is not changing much (only
last digit changes), which in comparison to frozen XY-plane, where there is
huge movement above the slab and within is rather ok to me. Unfortunately, I
have the issue with narrowing of the box in the unconstrained XYplane
simulation, which also destroys any attempts for PMF as I have mentioned
before.

> Also, it is worth comparing in single
> and double precision.

I will try this either.

> Still, I am not sure that you actually did a proper comparison for the
> following 2 reasons:
> 1. you have apparently different masses for the pulled group:
> Pull group 1:    15 atoms, mass   194.194
> Pull group 1:    15 atoms, mass   146.146

The difference is not much in my opinion, while molecules tested are similar
in size and are run in DOPC ok both of them.

> 2. you used very different starting depth offsets: -2.81855 is not
> very close to -1.60929

The difference is in the size of OCT slab (thick) and DOPC (narrow) and the
trouble is in the vicinity of the water phase, so the distances are (and
should be) different.

> PS: I would personally prefer if you avoided jumping directly to a bug report
unless you are sure of it. There is always at least the
> possibility that you are not using the code correctly.

Certainly and I appreciate two directions you gave me to try - at least I
will look inside octanol topology since I did not used octanol before our
tries of logP. However, I did not say anything about bug (I even searched
for term in my mail and I did not find it)

But how to set the box properly then?

> Chris.

Thank you both